From c451b1749bf304ff57dd629fb3fd92efb41e4d40 Mon Sep 17 00:00:00 2001
From: jhaveri-bhavya <jhaveri.bhavya.s@gmail.com>
Date: Sun, 6 Sep 2026 11:14:49 +0200
Subject: [PATCH] examples: show full Modelfile surface in create.py, add
 create-gguf.py

create.py only set from_ and system, which makes it look like create()
can't express a full Modelfile. Expand it to cover template, parameters,
messages and license too, with a comment mapping each kwarg to its
Modelfile directive.

Add create-gguf.py for the weights-import path: upload the file with
create_blob() and pass the returned digest through files=.

Refs #664, #431
---
 examples/README.md      |  3 ++-
 examples/create-gguf.py | 23 +++++++++++++++++++++++
 examples/create.py      | 17 ++++++++++++++++-
 3 files changed, 41 insertions(+), 2 deletions(-)
 create mode 100644 examples/create-gguf.py

diff --git a/examples/README.md b/examples/README.md
index 1df713ea..57ae1fe2 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -108,9 +108,10 @@ Requirement: `pip install tqdm`
 
 - [pull.py](pull.py)
 
-### Ollama Create - Create a model from a Modelfile
+### Ollama Create - Create a model from a base model or a GGUF file
 
 - [create.py](create.py)
+- [create-gguf.py](create-gguf.py)
 
 ### Ollama Embed - Generate embeddings with a model
 
diff --git a/examples/create-gguf.py b/examples/create-gguf.py
new file mode 100644
index 00000000..807c6cf6
--- /dev/null
+++ b/examples/create-gguf.py
@@ -0,0 +1,23 @@
+# Import a local GGUF file as a model.
+#
+# Modelfile equivalent:
+#   FROM ./model.gguf
+#
+# The API takes blob digests rather than file paths, so each file is uploaded
+# first with Client.create_blob(), which streams the file to the server and
+# returns its 'sha256:...' digest. The same pattern works for adapters=.
+
+from ollama import Client
+
+client = Client()
+
+path = 'path/to/model.gguf'  # replace with a real GGUF file on disk
+
+response = client.create(
+  model='my-gguf-model',
+  files={'model.gguf': client.create_blob(path)},
+  # quantize='q4_K_M',  # optional: quantize the weights during import
+  system='You are a helpful assistant.',
+  stream=False,
+)
+print(response.status)
diff --git a/examples/create.py b/examples/create.py
index 4ed8376f..d4c0e7b0 100755
--- a/examples/create.py
+++ b/examples/create.py
@@ -1,10 +1,25 @@
 from ollama import Client
 
 client = Client()
+
+# Each keyword argument maps to a Modelfile directive; together they are the
+# Modelfile, expressed in Python:
+#   from_      -> FROM        template   -> TEMPLATE     messages -> MESSAGE
+#   system     -> SYSTEM      parameters -> PARAMETER    license  -> LICENSE
 response = client.create(
   model='my-assistant',
   from_='gemma3',
-  system='You are mario from Super Mario Bros.',
+  system='You are Mario from Super Mario Bros.',
+  template='{{ .System }} {{ .Prompt }}',
+  parameters={'temperature': 0.6, 'num_ctx': 4096, 'stop': ['<end_of_turn>']},
+  messages=[
+    {'role': 'user', 'content': 'Who are you?'},
+    {'role': 'assistant', 'content': "It's-a me, Mario!"},
+  ],
+  license='MIT',
   stream=False,
 )
 print(response.status)
+
+# To import model weights from a local GGUF file instead of deriving from an
+# existing model, see create-gguf.py.